[r-t] A new Spliced Surprise Major canon

[Mark Davies]

Philip writes,

 > Did you start with your six chosen methods from the very outset?

No. I knew I wanted to start with Bristol, to use the 5056 no.1, and I 
knew I wanted methods that were (a) very good in their own right, but 
also (b) well-known. I set up a shortlist, which included other stuff 
like Yorkshire, Ashtead, Adelaide, Cat's Eye, Eleuthera, Heptonstall, 
Lancashire and Xenon. I then carried out many exploratory searches to 
find out which methods worked best.

In the earlier stages, in particular 2- and 3-spliced, it proved very 
hard to find methods which would give competitive music. At the other 
end of the scale, it started to become more difficult to achieve ATW. 
But I am pleased with the final set of methods - they meet all the 
objectives, whilst remaining a diverse bunch with a good spread of 
backworks and leadhead orders.

 > If I gave you, say, six "randomly" selected treble-dodging major
 > methods (or 6 of the same leadhead group you used, or
 > even slight tweaks like replacing Lessness with London (not that I'd
 > advocate this!)), what are the chances of a true atw composition
 > dropping out? With similar musical properties to your comps?

It probably depends on the cross-falseness and the leadhead groups, but 
ATW is the most likely property to be achievable. However, you'd 
probably need to find a different seed composition, and you're unlikely 
to get anywhere near the same music counts. I think I've ended up with a 
stand-out selection of methods, that go well together, and allow good 
music to be exploited. OK, I'm sure you could make improvements with 
methods deliberately constructed to optimise the music - but they might 
not end up as interesting to ring.

 > Is 6 methods the maximum you could get into an atw composition on this
 > specific calling plan? Your message kind of implies you've tried to
 > increase this?

No, I'm hopeful of adding further methods in. However with the current 
version of the software I'm hitting some memory problems - the 
cross-falseness tables can take 8GB of memory and more by the time I 
reach 7 and 8 methods. I've got 16GB in my main desktop, but the 
table-build time is a bit excessive, and its easy to run into GC or 
page-faulting problems which kill the searches, especially if you want 
to run more than one.

Since my modus operandi is based around relatively quick searches, which 
I chain together after visually inspecting the results and tuning 
parameters for the next iteration, the memory requirements and 
table-build times start to become onerous over six methods. I'm working 
on some improvements which should alleviate this; I've also got a 
prototype which is meant to "double-up" the ATW of a single method, in 
preparation for replacement of half its leads by a new method during a 
second phase, which should provide another avenue to higher 
multi-spliced stages.

 > the new thing here is the series of methods with constraint of a fixed
 > calling. I get a bit uneasy about constructed approaches as you're
 > essentially condemned to a damage-limitation exercise

Well, I'm guessing the stochastic approach is new, although happy to be 
corrected on that front (someone gave me the idea for applying simulated 
annealing to spliced nearly twenty years ago, so it ought to have been 
exercised before now!). You are right, I am trying to optimise the 
result around a fixed framework, but I rather like that - it's the soul 
of ringing really, isn't it?

> ...how applicable is your general metaheuristic approach to other more
> "architectured" ringing problems, where say the calling may be fixed but
> instead the method choice is open?

Very much so, although the software would need to have major 
adaptations. You're looking at searching place notations not spliced 
callings.

> 1) Getting methods for a whole-course 23-spliced major, along the lines of
> what Richard Smith did at:  ...

Definitely, although looking at Richard's post it seems he did use some 
kind of stochastic algorithm. I don't know by what extent it would be 
possible to improve his on work.

> 2) Starting with say Alan Reading's recent 8-part major with all the runs,
> knowing you wanted to replace 3 identified methods with replacements that
> would leave the composition true and with the same musical properties

Yep, definitely. Again, you'd need different software - as above, I'm 
guessing it would be more a case of exploring the space of place 
notations that make up the three replacement methods.

> 3) Producing an extent of spliced treble dodging major? (Either
> "conventional" spliced, or 180 whole courses, or a 7-part with 180 leads
> per part)

You can have one of those once you've rung the Helixoid/PB extent. :-D

MBD